Hyperspectral Data Processing: Algorithm Design and Analysis

10

Random Endmember Extraction Algorithms (REEAs)

The initialization-driven endmember extraction algorithms (ID-EEAs) developed in Chapter 9 intend to address the issue arising in the use of random initial endmembers, which causes inconsistency in final results. Interestingly, this disadvantage can become an advantage if we look at this issue from a different point of view. This chapter presents a rather different approach that allows an EEA to take advantage of the randomness to produce a set of desired endmembers. The idea is to implement an EEA using random initial endmembers as a random algorithm where a single run of a random EEA is defined as a process of running an EEA using a set of random initial endmembers and the results produced by a single run are considered as a realization. If there is an endmember, it should appear in realizations regardless of what random initial endmembers are used. In light of this interpretation, taking the intersection of realizations eventually converges to a common set made up of the desired endmembers in which case an EEA is terminated and the number of endmembers is then automatically determined by this set without appealing for any criterion. An EEA implemented in such a manner is called random EEA (REEA).